Critical use of GenAI: Creation

Within education, outright creation is probably the most controversial use of LLMs.

After all, if we ask students to create something, we're usually interested in having them experience the brainstorming, the dead ends, the struggle of putting an idea to practice and the satisfaction of seeing things come together at the end of this (arduous) process. The learning is happening during the journey, so why take that away from students?

Truth is, there are no good reasons to do so, but there is a real risk that students self-sabotage and skip over the process itself. Their reasons will vary from self-doubt to time pressure to just not really enjoying the friction that comes with learning. As a teacher, you can counteract these tendencies by being upfront with students about the value of struggle during ideation. You can also make sure that you monitor or even assess their process, to put the incentives in the right places.

Alternatively, you can allow them to use LLMs for creation while emphasising critical thinking, which is the topic of this post. Building on the tentative framework for critical thinking about GenAI, here we will look at the creation of artefacts or ideas, the subcategories in Brachman's ontology (Brachman et al., 2024).

Artefacts

Artefacts are things that can be used directly or after some modification, such as research proposals, intervention strategies and prototype descriptions. The five-step process model of thinking applies to the development of such artefacts, with all design choices being 'action selections', ideally consequences of a critical reflection process. The risk of GenAI use is that the action selection happens without such reflection, so in teaching you would have to ensure that students go through all the steps.

Problem identification

Any artefact is supposed to meet a need. Regardless of the assignment you give students, it makes sense for them to think about what the artefact is supposed to accomplish. This means:

• Understanding the problem that needs fixing
• Identifying what counts as solving the problem: when can you say an artefact is good?
• Conversely, identifying what failure would look like

Information gathering

Now that the problem is known, students can explore what has already been done to tackle it.

• Which earlier attempts were made? What were they based on?
• How well did they work? What were their shortcomings?

Whether it's reading journal articles to prepare for a research proposal, comparing marketing strategies and exploring their rationales or putting technical specifications side by side, this step is an exploration stage.

Sensemaking

In the sensemaking step, the students write down criteria for artefact quality, based on their findings from the earlier steps. These evaluation criteria will ultimately be used to judge GenAI output, but also to properly prompt the LLM in the first place.

Belief formation

Belief formation is a refinement of the previous step, in which students take a stance on the relative importance of the different criteria. Here, they make clear decisions on particular features that should be present in the artefact.

Action selection

Only a this point can the students prompt a GenAI system to generate an artefact. The prompt can now be clear in its requirements and, more importantly, the students can evaluate the GenAI quality on the basis of their own benchmarks. This allows them to refine the artefact through iterative prompting or modification

Ideas

Ideas differ from artefacts in that they are insights that can give direction to a thought process or creative process. With artefact generation, the GenAI output, once critically assessed and properly revised, is the goal. When generating ideas, the GenAI output is often the start of something else.

Of course, to critically assess the value of an idea, similar logic applies as was the case for artefacts. You need to have a good notion of what makes an idea a good idea before you can evaluate it. This is done by problem identification (what is the purpose of the idea?), information gathering (what relevant knowledge is out there?) and sensemaking (which knowledge should be part of the idea?). However, it's already at the step of belief formation that students invoke GenAI, using the technology to brainstorm what should go between the broad knowledge they obtained and setting up a subsequent course of action.

To think critically here, the challenge for students will be to generate multiple ideas and pit them against each other using some systematic approach. This could again be benchmarks, but approaches like an impact/feasibility matrix might also work. In any case, the critical step will again be to decide whether GenAI output is good for the purpose at hand.

Importance of preparation

As these two approaches show, critical thinking during creation with GenAI is strongly dependent on the work done before the prompt. The better the preparation, the deeper the considerations can be once GenAI starts spewing results. Teachers might opt to include an extra step in which students sketch an artefact or create an idea without GenAI first, so that they have some human ideation to add to their overall analysis. In any case, the critical thinking is unlikely to happen without due preparation, shifting the focus of teachers to ensuring a good process. As the next posts will show, the other categories in Brachman's ontology also require a more process-oriented mode of teaching.

References

Brachman, M., El-Ashry, A., Dugan, C., & Geyer, W. (2024, May). How knowledge workers use and want to use LLMs in an enterprise context. In Extended Abstracts of the CHI Conference on Human Factors in Computing Systems (pp. 1-8).